Bug Report

Summary

RelativePath.GetHashCode uses a custom vectorised function for calculating the HashCode of a string that I once made for my Virtual FileSystem. This function is case-sensitive (as it hashes the raw bytes), however the current specification suggests that paths should be compared case insensitive.

More precisely, our GetHashCode is case sensitive, while our Equals is case insensitive, which breaks the contract.

Steps to reproduce

Hash two RelativePath(s) with different casing, and note the hash differs.

What is the expected behaviour?

For any two strings which are equal with Equals, their GetHashCode must match too.

Design

Apply the following steps:

Create helper method which converts to lower on stack, before hashing. GetNonRandomizedHashCode32Lower
Create benchmark for Hashing with string.GetHashCode(OrdinalIgnoreCase) vs GetNonRandomizedHashCode32Lower
Compare the performance of the two, pick the faster of the two.
Apply the new faster hash function to AbsolutePath too.

Note:

Use TextInfo.ChangeCase<ToUpperConversion> for changing the case. It contains a backported vectorised implementation from .NET 8. We can switch to regular ToLower Span extension once we start targeting .NET 8. If you want, can use #if !NET7_0_OR_GREATER define to include conditional, so we don't forget to replace it.

Other information

When the Path library was initially rewritten by me, the requirements and implementation were a bit different. Back then we converted RelativePath(s) into lowercase. However, circumstances have changed, and we stopped doing this a while back.

Nexus-Mods / NexusMods.Paths

Bug + Perf: RelativePath.GetHashCode is Case Sensitive #25